Adversarial learning has emerged as a unifying paradigm across machine learning, security, perception, and complex socio-technical systems, particularly in environments characterized by noisy labels, weak supervision, and strategic manipulation. This article develops a comprehensive and theoretically grounded synthesis of adversarial learning under noise, drawing strictly from foundational and contemporary literature spanning noisy example learning, adversarial label learning, generative adversarial networks, weak supervision, and adversarial robustness in applied domains such as network intrusion detection, medical signal analysis, and urban traffic systems. The study advances an integrated conceptual framework that treats noise, adversarial behavior, and supervision uncertainty not as isolated challenges but as structurally related phenomena that shape learning dynamics. Through extensive methodological exposition, the article explicates how stochastic adversarial labels, weak supervision frameworks, and adversarial training objectives interact with distributional distances, transparency mechanisms, and robustness constraints. The results are presented as a detailed descriptive synthesis of theoretical and empirical findings reported in the literature, emphasizing patterns, trade-offs, and emergent properties rather than numerical outcomes. The discussion critically examines limitations in current adversarial learning approaches, including scalability, interpretability, and domain transferability, while outlining future research trajectories that bridge probabilistic learning theory, adversarial security analysis, and real-world deployment. By offering an exhaustive elaboration of adversarial learning under noise, this work contributes a publication-ready reference that consolidates fragmented insights into a coherent methodological and conceptual foundation for robust machine learning in adversarial environments.